OperationKit

mcp
Security Audit
Fail
Health Warn
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 6 GitHub stars
Code Fail
  • Hardcoded secret — Potential hardcoded credential in app/client/src/components/AccountSettings.apikey.test.tsx
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

AI transformation infrastructure. A self-hosted knowledge base, agent board and MCP connection for putting AI to work inside a business. Run it in-house, license it as a consultant, or have us set up your server and run it for you. Works with Claude, Codex, Grok or open-weight models. Apache-2.0.

README.md

Operation Kit

The infrastructure layer for AI implementation consultants.

Install it once per client, on the client's own server. They own it; you build on it.

Operation Kit is a ready-made base for putting AI to work inside a business. A consultant installs
one copy per client, on a server the client owns, in under an hour. The client gets a private web
app with a knowledge base of their own documents, AI workers that run jobs as cards on a board, an
optional chat with ready-made agents, and a secure connection for the AI apps their people already
use (Claude, ChatGPT, Grok). The consultant starts every engagement 80 to 90 percent of the way
there and spends their time on the client's actual workflows.

Why the client owns the install

  • Their server, their data. Every client has its own copy on its own server. There is no shared
    Operation Kit cloud and no central database holding several clients' data.
  • Their AI accounts. Work runs on the client's own Claude subscription or API key (OpenAI and
    Gemini optional). No pooled or resold seats.
  • Fewer vendor-risk questions. Because client data never sits on a vendor's systems, the usual
    "is your vendor SOC 2 audited?" question mostly becomes "is our own server set up well?". To be
    plain about it: Operation Kit has no SOC 2 attestation, and we do not claim one. See the
    security overview for what is protected and what is not done yet.
  • No lock-in to a model. Agents, skills and the knowledge base are plain files on the server.
    Switch the AI provider and none of it moves.

What is inside

What it does
Knowledge base Drop in Excel, CSV, PDF and Word files; they become organized, linked notes you can ask questions of, with every answer citing its source. It works like a living wiki: originals are kept untouched, a page per person, organization and topic is compiled from them, good answers can be saved back as new pages, and a nightly cleanup flags contradictions and stale facts. Claude and ChatGPT chat exports and call transcripts (.txt, .docx, .vtt, .srt) can be imported too. Plain markdown, so it opens in Obsidian or any editor.
AI workers on a board Anything that takes more than a few minutes becomes a card: queue → working → review → done. Each card runs a coding agent (Claude Code by default; Codex and Grok supported) in its own sandbox, and an independent AI reviewer checks the result before a person signs it off.
Agent chat (optional) One ongoing chat per agent, like a personal assistant. Agents read and update their own notes, draft emails and calendar invites for a person to send (never on their own), and hand bigger jobs to the board.
Connect any AI app (MCP) Claude, ChatGPT, Grok, Codex, Cursor and other MCP apps connect with a named, limited key and can use the same knowledge base, agents and board. Connect your AI.
Agent templates Each agent is one file (instructions, knowledge folders, tools, schedules). Starter templates ship in the box (Personal assistant, Knowledge librarian, Operator); consultants can write their own and drop them into a client's install. A shared marketplace for publishing agents is planned, and the install side is already built for it (how it would plug in).
Connectors Link GitHub and Google Workspace once in Settings and choose which agents may use each. Credentials are stored encrypted and never shown to agents.
Workspaces Separate teams or business units inside one install, each with its own board and knowledge folders.

Install

A fresh Ubuntu 24.04 server, the client's domain on Cloudflare, and a Claude account. Claude Code on
your laptop follows INSTALL.md and does the work; you answer a few questions and do two
short sign-ins. About 30 to 45 minutes, most of it waiting for the build.

The code: https://github.com/m1keluka/OperationKit (updates come from the same place, and are
applied by hand: Updating).

Recommended stack (what we run and test): an OVHcloud Rise dedicated server, Backblaze B2
for encrypted off-site backups, and Cloudflare (Free) for the domain, the tunnel and the sign-in.
docs/RECOMMENDED-STACK.md walks through it end to end, from ordering
the server to GitHub tokens and updates. Any Ubuntu 24.04 server works.

How big a server. Plan 6 GB of memory plus 3 GB per AI session running at once, 2 CPU threads
plus 1.5 per session, and 60 GB of disk plus about 3 GB per session plus the client's files:

Sessions at once Server Example (prices checked 2026-11-01)
1-2 (solo) 8+ threads, 16 GB, 150 GB OVH RISE-1 from $64/month
3-5 (small team) 12+ threads, 32 GB, 250 GB OVH RISE-2 from $72/month
6-10+ (heavy) 24-32 threads, 64-128 GB, 500 GB+ OVH RISE-4 from about EUR 160/month

Measured numbers, alternatives and the reasoning: choosing a server.

 Person ──► Cloudflare Access ──► Cloudflare Tunnel ──► client's server (no open web ports)
            who are you?          private connection    ├─ Operation Kit app
                                                        └─ one sandbox per AI job

Docs

For Read
Setting up from nothing: server size, OVH, Cloudflare, B2 backups, seats, GitHub, updates docs/RECOMMENDED-STACK.md
Installing for a client INSTALL.md
Connecting Claude, ChatGPT, Grok or another AI app docs/CONNECT-YOUR-AGENT.md
"Where does our data go?" (plain-language, one page) docs/SECURITY-OVERVIEW.md
Running a live demo docs/DEMO-RUNBOOK.md
Using it day to day docs/product/README.md
Agent chat docs/product/09-agent-chat.md
Writing agent files docs/agents/agent-file-format.md
Agent / HTTP / MCP API reference docs/api/README.md
Model choice and portability docs/product/08-model-portability.md
Engineers: architecture and threat model docs/architecture/README.md · SECURITY.md

Status

Built to be a solid self-hosted install: Cloudflare in front, no open web ports, sandboxed AI jobs,
encrypted secrets, a forced password change for the first admin, encrypted off-site backups with a
restore test, and automatic security updates for the operating system (Ubuntu's patches; Operation
Kit itself is updated by hand, see Updating). It is not an audited SaaS. Known gaps (no SOC 2, no built-in two-factor or SSO beyond Cloudflare Access,
transcripts not encrypted at rest) are listed in the security overview.

License

Apache-2.0, see LICENSE.

Reviews (0)

No results found